Josef Fruehwald
18 October 2017
Vowels are mostly articulated with the tongue body.
Vowels can be largely described in a two dimensional space.
These tongue shapes have characteristic acoustics.
Some vowels have complex articulations.
If you know where in a recording a vowel is…
…you can measure them
- Field Site A
- Speaker A-1
- Speaker A-1 audio
- Speaker A-1 annotation
- Speaker A-1 data
- Speaker A-2
- ...
- ...
- Field Site B
- ...
library(purrr)
library(dplyr)
# Glob for file names
filenames <- Sys.glob("data/PH*/PH*/*meas.txt")
# Define the reading function
read_iy <- function(filename){
dat <- read.delim(filename)
out <- subset(dat, plt_vclass == "iy")
return(out)
}
# Map
iy_df_list <- map(filenames, read_iy)
# Combine
iy_df <- bind_rows(iy_df_list)There are a few desiderata for modelling vowel shifts like these:
Inter-speaker differences.
Generalized Additive Models in the mgcv package fit all of these desiderata.
The non-linear models fit with gamms utilize various flavors of “splines.” They all have slightly different mathematical properties, but there are a few things they have in common. First, they all involve some “basis”.
Each of these polynomials are then weighted.